How often should you refresh Meta ad creatives, and how many should you test per week?

What are the signs of creative fatigue on Meta?

Creative fatigue shows up as three things moving together: frequency rising, click-through rate falling against the ad's own baseline, and cost per acquisition climbing. One 2026 guide sets the cold prospecting watch line at a 7-day frequency of 2.5 and calls a CTR drop of 20 to 30% from the creative's own baseline over 3 to 7 days the arrival of fatigue, while noting retargeting can tolerate a frequency of 5 to 7 (Superads). Another agency guide uses 2.5 to 3.0 as a soft warning for prospecting and 3.0+ as the point to refresh (inBeat).

Compare the ad to itself, not to the account average. A static that launched at a 1.2% CTR and now runs at 0.9% is fatiguing even if 0.9% beats the account mean.

If CPA is climbing without that pattern (frequency flat, CTR holding), fatigue is probably not the cause. Our guide on the other causes of a rising Meta CPA covers auction price, budget drift, click quality and conversion.

Do most Meta ads actually reach creative fatigue?

Most do not. A 2026 benchmark of 368 ads across 15 DTC brands (48.5 million impressions, April to August 2026) found that 80% of creatives never reached 100,000 impressions, the median creative ran 18 days with 10,665 lifetime impressions, and the top 10% of creatives carried 68% of spend (The Interconnections benchmark).

For the creatives that did scale, the same benchmark reports CTR dropping about 5.5% early and then flattening, CPM staying roughly flat, and CPA rising 19.6% by the 500,000 to 1 million impression mark (The Interconnections benchmark).

That changes the job. If the delivery system concentrates spend on a few winners, the real constraint is hit rate: how many net-new concepts you need to ship to find the next ad that can carry spend. Meta's retrieval system, Andromeda, narrows tens of millions of ad candidates down to a few thousand relevant ones per request (Meta Engineering). Ten small variations of one idea give that system very little new to choose from.

How often should you refresh Meta ad creatives?

Plan fresh creative every 1 to 2 weeks on high-spend, narrow-audience campaigns and every 2 to 4 weeks on lower-spend or broad campaigns, then let the fatigue signals above override the calendar (Superads). inBeat frames the same idea by audience size: every 7 to 10 days under 100,000 people, 10 to 14 days at 100,000 to 500,000, and 14 to 21 days above that (inBeat).

In practice we separate two jobs. The refresh job keeps proven concepts alive with new hooks, new first frames and new formats (a winning UGC script cut as a static, a static rebuilt as a carousel). The discovery job ships net-new concepts: a new angle, a new mechanism, a new avatar. Refreshes protect this month's spend. Discovery protects next quarter's.

How many new creatives should you test per week at your spend level?

Published guidance ranges from about one concept per $2,000 to $3,000 of monthly spend down to a handful per month per $50,000, so start from the ranges below and adjust to your production capacity and hit rate. MHI Growth Engine gives tiered targets and a baseline of one new concept per $2,000 to $3,000 in monthly Meta spend (MHI Growth Engine).

Source: MHI Growth Engine. A different 2026 guide recommends three to five genuinely new concepts per month for every $50,000 of monthly spend, meaning different hooks and angles rather than variations (The Interconnections). The gap between those two numbers is mostly about what counts as a "concept". The lower figure counts only genuinely new ideas. The higher one includes new executions.

Our rule: count concepts, not files. Five headline swaps on one image are one concept. A new angle (say, sleep quality instead of energy for the same magnesium product) with its own hook, visual and landing page is a new concept.

How much should you spend on each creative test before judging it?

Spend 2 to 3x your target CPA per creative, with a floor around $100 to $150, before judging it on purchases (SparkUGC). Some frameworks go higher, at 3 to 5x target cost per purchase and a minimum of 7 days to cover day-of-week swings (Admade). MHI Growth Engine splits it by what you want to learn: $100 to $200 to validate the hook and CTR, $300 to $500 for add-to-cart and checkout signals, and $500 to $1,000+ to read purchase CPA with confidence (MHI Growth Engine).

Keep Meta's own learning threshold in mind: an ad set needs a minimum of 50 optimization events over 7 days to stabilize, and significant edits can restart learning (Meta for Developers). A single creative test rarely reaches that alone, which is why we judge early tests on attention metrics and let purchase data accumulate at the concept level.

When should you kill a Meta ad creative?

Kill early on attention and late on cost. Cited kill rules include: zero purchases after spending 3x target CPA, CTR under 0.5% after 1,000 impressions, or CPC at 3x the account average with no purchase signal (Admade). SparkUGC adds a hook-rate rule (under 20% after about 2,000 impressions) and warns not to kill on CPA in the first 48 hours because fresh-creative CPA is the noisiest signal (SparkUGC).

Our kill and promote checklist

1. Day 1 to 2: ignore CPA. Check that the ad is spending and that hook rate (video) or CTR (static) is not far below your account norm.
2. After about 1,000 to 2,000 impressions: kill on weak attention (CTR under 0.5%, hook rate under 20%).
3. At 3x target CPA with zero purchases: kill.
4. Beating the control by at least 15% on CPA with 3+ conversions: promote to the scaling campaign, per SparkUGC's threshold.
5. Everything in between: let it finish its planned budget and read it at the concept level, not the single-ad level.

How we plan creative volume and refresh supplement ads at Succession

We plan creative as a system, so fatigue becomes a scheduling question instead of an emergency. This is the order we use to supply both the refresh job and the discovery job each week.

1. Research mechanisms and angles before writing briefs. For a supplement that means the ingredient's mechanism in plain language, the angles buyers already respond to, and the angles competitors have worn out.
2. Ladder sub-avatars. One core product gets several specific sub-audiences, each with its own hook (for a hypothetical magnesium formula: shift workers, new parents, people who train hard). Each sub-avatar is a source of net-new concepts, not just new files.
3. Run every concept through our Schwartz gates: mass desire, awareness level, market sophistication, headline, body. In crowded supplement categories, sophistication is the gate that fails most often: the claim has been heard, so the concept needs a new mechanism or angle to be believed.
4. Write compliance-first. Structure/function language, no disease claims and no condition attributed to the viewer, checked before production so a strong concept is not lost to a rejection.
5. Build the landing path with the ad. Message match from hook to headline, and an advertorial or listicle for colder, less aware sub-avatars (see whether advertorials beat product pages).
6. Read tests at the concept level, and read the page too. Next to CPA and ROAS we look at conversion rate and revenue per session, so a good ad sent to a leaking page is not killed as a bad ad.

Refreshes (new hooks, first frames and formats on proven concepts) keep this month's spend alive. Research, laddering and the gates keep the discovery job supplied. If you are not sure which awareness levels your current ads cover, start with Schwartz's awareness levels applied to Meta ads.

What does this look like for a $50k a month supplement account?

Here is a worked example with hypothetical numbers, to show how the pieces fit. Assume a supplement brand spending $50,000 a month on Meta with a $40 target CPA.

The math is easy. Production is the constraint: five genuinely different concepts a week means research, scripting, shooting or designing, compliance review and a matching landing page angle every week. That is why we plan creative volume and landing pages together. A new angle that sends traffic to a page written for a different angle wastes the test.

If you want us to size this for your account, book a free growth audit or see our services. More operator guides are in guides.

FAQ